Skip to content

[WIP] Add Polygon model support #837

Closed
Om-Doiphode wants to merge 3 commits intoweecology:mainfrom
Om-Doiphode:polygon_support
Closed

[WIP] Add Polygon model support #837
Om-Doiphode wants to merge 3 commits intoweecology:mainfrom
Om-Doiphode:polygon_support

Conversation

@Om-Doiphode
Copy link
Contributor

@Om-Doiphode Om-Doiphode commented Nov 16, 2024

Fixes issue #758

  1. Add MaskRCNN to deepforest/models/ directory.
  2. Create a polygon dataset class.
  3. Confirm that preprocess.split_raster and other utilities/visualization functions can take in polygon predictions.
  4. Create a example ipython notebook showing how to train a polygon model, train an initial model using open source polygon data

@bw4sz bw4sz added this to the DeepForest 2.0 milestone Jul 3, 2025
@bw4sz bw4sz removed this from the DeepForest 2.0 milestone Oct 2, 2025
@ethanwhite
Copy link
Member

Thanks for the contribution on this @Om-Doiphode! The code base has moved on in the past year and we've been doing some of this work elsewhere, so I'm going to go ahead and close this. @Om-Doiphode if you still want to work on this let me know and I can link you up with the folks currently working in this space

@ethanwhite ethanwhite closed this Mar 12, 2026
@Om-Doiphode
Copy link
Contributor Author

Yes, I would like to work on this.

@ethanwhite
Copy link
Member

Thanks @Om-Doiphode

@jveitchmichaelis - @Om-Doiphode work on polygon support for us in Fall 2024 and then we lost track of it. Can you catch him up on where things are and see if he can be of help in the work you've been doing to generalize everything to points and polygons?

@ethanwhite ethanwhite reopened this Mar 17, 2026
@jveitchmichaelis
Copy link
Collaborator

Hi @Om-Doiphode,

First thing would be to align the polygon dataloader with #1311 (ie inherit from the base dataset, modify only the bits that need adjusting). We would also need to check what settings the augmentation pipeline needs. And then add unit tests for that. I think we probably also need to look into our utilities for to/from COCO format.

The evaluation code should already work with polygons, but we don't have any testing in place yet.

Once these two are in place, we can look at model training. It's easier that we have the eval stage working first, because we can sanity check using any dataset (e.g. MS-COCO). I will think about a good training dataset, but probably something like OAM-TCD as I'm familiar.

@Om-Doiphode
Copy link
Contributor Author

Since this branch is way behind the main branch (338 commits behind), should I go with git merge main or git rebase main?

@jveitchmichaelis
Copy link
Collaborator

I would either rebase or start from scratch and copy your changes in.

@Om-Doiphode
Copy link
Contributor Author

Om-Doiphode commented Mar 24, 2026

I have issued a new PR here: #1359
This PR is still a WIP and I will complete it ASAP.

@bw4sz
Copy link
Collaborator

bw4sz commented Mar 25, 2026

this is superceded by 1359

@bw4sz bw4sz closed this Mar 25, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants